Once AI spend crosses a few thousand dollars a month and multiple teams are using it, the question shifts from 'is this worth it' to 'whose budget does this come out of.' Chargebacks and showbacks are the two standard answers, and picking the wrong one for your business size creates friction that outweighs whatever accounting tidiness it was meant to deliver.
The difference, in practice
Showback: teams see their own AI usage and cost, but it's still paid from a central budget -- visibility without financial consequence
Chargeback: a team's AI spend is actually deducted from their own departmental budget, creating a direct financial incentive to use it efficiently
Showback is lower friction to set up and rarely triggers political pushback; chargeback drives more disciplined usage but needs real buy-in to avoid resentment
Which one fits which size of business
For a business under about 40 staff with AI usage still concentrated in one or two functions, showback is almost always the right starting point. The administrative overhead of running a full chargeback system, allocating costs precisely, handling disputes about which team's workflow triggered which spend, usually costs more in finance-team time than it saves in disciplined usage at that scale. A Brisbane engineering firm we worked with tried chargeback at 25 staff and abandoned it within a quarter, reverting to showback, after finding the allocation disputes between two overlapping project teams consumed more finance-team hours monthly than the AI spend itself.
When chargeback earns its overhead
Chargeback starts paying for its own complexity once a business has genuinely separate cost centres with their own P&L accountability, typically past 60 to 80 staff, or in a professional services firm billing AI-assisted work back to specific clients, where the allocation isn't arbitrary, it maps directly onto real client billing. In that setting, chargeback isn't overhead, it's simply accurate accounting that was already needed for other reasons.
A practical middle path
A workable middle ground for a growing business: run showback as the default, but flag any team whose usage grows disproportionately for a direct conversation rather than an automatic charge. This catches genuine cost problems (a team running an inefficient workflow, or one that's scaled far beyond its original use case) without building a full chargeback bureaucracy before the business is actually big enough to need one.
If you're trying to work out which model fits your business as AI usage spreads across teams, get in touch through /contact and we'll help you think through the trade-offs for your specific structure.
What the numbers looked like at the Brisbane firm
Before abandoning chargeback, the Brisbane engineering firm was spending roughly eight hours a month of a finance coordinator's time resolving allocation disputes between two project teams whose work genuinely overlapped on a shared AI-assisted document review workflow. At a loaded rate of around $55 an hour, that's about $440 a month spent adjudicating who owed what, against a total AI spend across both teams of only $610 a month. The overhead of running chargeback was consuming nearly three-quarters of the value of the thing it was meant to be tracking, which is exactly the trap a smaller business needs to watch for before adopting a system built for a much larger organisation.
Setting expectations before rolling either model out
Whichever model you choose, tell teams upfront how their usage will be visible and, if applicable, charged, before they start using the tool rather than after. A team that discovers months into using an AI workflow that its spend has quietly been tracked and is about to become a chargeback line item tends to react defensively, even when the actual dollar amounts are modest. A five-minute conversation at rollout, explaining the model and why it was chosen, avoids most of the friction that otherwise surfaces later as resentment about a system nobody explained properly at the start.
A Sydney-based professional services firm billing AI-assisted research back to specific client matters found chargeback straightforward precisely because the allocation logic already existed: time was tracked per matter regardless of whether a human or Claude did the work, and AI cost simply became another line in an existing client-billing system rather than a new administrative structure built from scratch.
Whichever model a business starts with, it's worth revisiting the choice once total AI spend crosses roughly $50,000 a year across the organisation, because that's typically the point where the accuracy benefits of proper cost allocation start to outweigh the administrative overhead, even for a business that was too small for chargeback to make sense a year or two earlier.
A useful rule for picking between them in the meantime: if you can't cleanly answer which team caused a given dollar of AI spend without a meeting to work it out, you're not ready for chargeback yet, no matter how large the total number has grown.
